Prithviraj (Raj) Ammanabrolu
@rajammanabrolu
Followers
7K
Following
15K
Media
405
Statuses
3K
Interactive & grounded AI, RL, Language. Assistant Prof @UCSanDiego. Research Scientist @Nvidia.
San Diego, CA
Joined April 2019
The PEARLS Lab at @UCSD_CSE is now open for business! I'm recruiting Fall 24 PhD students in all things interactive and grounded AI, RL, and NLP!! Join us in the land of 🏖️ beach (🧋pearl tea included). Apply by Dec 20. Please help spread the word!. More:
Soon™, I'll be an Asst Prof @UCSanDiego @UCSD_CSE focusing on interactive & grounded AI, RL, NLP. I will also be a research scientist @MosaicML helping lead efforts to make tech like RLHF more accessible. Looking for PhD students & research eng/scientists to join me in ☀️SoCal🏖️
7
65
250
RT @jaseweston: 🌿Introducing NaturalThoughts 🌿. 🎯 Data curation for general reasoning capabilities is still relativ….
0
71
0
I've been knighted and granted the legendary weapon: The Sword of the Mosaics. A worthy addition to my mantel. Thanks @jefrankle
0
0
18
I get a constant question "you have so many AI industry contacts, why do you bother keeping the Prof affiliation? 20x+ more salary, it's the time to be in industry". My only allegiance is to open research. If someone in my position doesn't work towards it, who else will.
4
37
126
Overlapping research interests between my UCSD lab and the industry side is really the only way to make sure I'm not doing two 100% jobs. Everyone I've spoken to top leadership down at Nvidia have impressed me with their commitment to open research! High alignment potential!!.
0
1
9
My next professional move is to go to the Source of the Compute. Soon™ I'll be hanging out with the incredible researchers @nvidia as a RS working on open source/science post training esp reasoning VLA models for embodied agents! There is no ASGI without embodiment!.
My next move was to touch grass. Would recommend.
8
5
104
On the topic of improving RL efficiency. Being able to reduce memory footprint is important for actor critic methods. More stable fp8 training is v exciting!.
Deep learning training is a mathematical dumpster fire. But it turns out that if you *fix* the math, everything kinda just works…fp8 training, hyperparameter transfer, training stability, and more. [1/n]
0
1
9
RT @davisblalock: Deep learning training is a mathematical dumpster fire. But it turns out that if you *fix* the math, everything kinda ju….
0
147
0
This looks like a slide I've been using in my class for a year or two. "The primary reason to improve MLSys Inference efficiency is for more RL Exploration". Inference speed and verification are the two main bottlenecks on which progress timelines depend.
So I think something else that doesn't get discussed much is the extrapolation of this inference : training trend. - 2015: back in the day, we would train one model per dataset, and inference it once (to obtain the eval result for our paper).- 2020: with chatgpt, multi-task
2
0
13
On behalf of every AI researcher who has negotiated offers in the last bit, I wanna say thank you to Meta. Bless y'all for moving the entire market baseline up.
As Mark Zuckerberg lures away top research talent to Meta, OpenAI executives say they're ‘recalibrating comp,’ according to an internal memo.
0
0
46
My next move was to touch grass. Would recommend.
3
0
39
RT @ChengZhoujun: 🤯What we know about RL for reasoning might not hold outside math and code?. We revisit established findings on RL for LLM….
0
55
0
For those in my DMs ;)
I recently left @DbrxMosaicAI. It's been a ride building out the RL team from <4 ppl to 20+ across two companies & acquisition +figuring out RL as a Service in prod. Mosaic had insane talent density. Some "relaxation" while I put out Prof fires for a smol bit then new adventures!
0
0
25
I recently left @DbrxMosaicAI. It's been a ride building out the RL team from <4 ppl to 20+ across two companies & acquisition +figuring out RL as a Service in prod. Mosaic had insane talent density. Some "relaxation" while I put out Prof fires for a smol bit then new adventures!
11
3
149
RT @mark_riedl: If you work in the intersection of NLP and games/narrative, then this workshop is for you! Organiz….
0
2
0
Wordplay is back! We'll be co-located with EMNLP this year in Suzhou. As always come hang out and chat everything language agents+ interactive learning!!.
0
2
9
The thing that feels so off about the core tech world is that every convo is very transactional. Maybe true elsewhere too. "Oh you're an expert in RL, can you answer questions about my new startup?". Every single (Bay) party. No I do not want to consult. I just wanna hang out.
2
0
31
RT @rm_rafailov: When we first published our work on this 9 months ago it was rejected for being impractical in realistic cases. Six month….
0
14
0
Cool thread
The solution to (1) is obvious: use a humanoid. Don’t overthink it. The human is a biological humanoid.
0
0
2
I'm kinda humanoid robot pilled now. Not sure when the shift happened but a while ago I was definitely "let's do specialized hardware per task". Ease of data collection and seeing results re scaling laws for embodied agents in my own lab and out has def changed my mind.
1
0
9
LLMs in Barbies. Didn't think @meetM3GAN 2.0 would be out in the real world already.
Mattel and OpenAI are teaming up. Their first AI powered toy will arrive later this year, just in time for Christmas. They are also incorporating OAI Enterprise company wide. No details on the toy yet, but if we get an AI Barbie expect some amusing jailbreaking.
2
1
7